A Novel Back-off Algorithm for Integration of Scheduling and Control of Batch Processes under Uncertainty

Abstract

A novel decomposition algorithm for the integration of scheduling and control of multiproduct, multiunit batch processes under stochastic parameter uncertainty is presented. This iterative algorithm solves a scheduling and dynamic optimization problem around a nominal point while approximating uncertainty through backoff terms, embedded in the operational process constraints. Monte Carlo simulations are performed to propagate uncertainty and to evaluate dynamic feasibility; statistical information is drawn from these simulations to update the back-off terms. Convergence of the algorithm results in a set of scheduling and control decisions that aim to keep the plant dynamically feasible under the effect of uncertainty up to a user-defined tolerance criterion. The proposed algorithm is shown to be successfully applied to a multiproduct, multiunit batch plant under the effects of different probability density functions in the uncertain parameters. The algorithm's performance is gauged against a fully integrated, mixed logic dynamic optimization problem with multiscenario-based uncertainty. The solution to the integrated algorithm is obtained with mixed integer nonlinear programming solvers. Results show that the proposed decomposition algorithm remains computationally attractive, without compromising the quality of the solution.

Integration of planning, scheduling, and control: A review and new perspectives

“…This approach has been applied to a multiproduct continuous system, including process design [ 83 ] and multiproduct, multiunit batch systems. [ 84,85 ] However, for large‐scale problems, the computational cost may be expensive, since the Monte Carlo simulations represent the most intensive calculations in the back‐off approach.…”

Safe Chance Constrained Reinforcement Learning for Batch Process Control

“…There have been two methods proposed recently, by [25,26], which integrate a similar penalty method into the RL problem properly, and achieve high probability constraint satisfaction. This is achieved through deployment of the concept of constraint tightening, which is common to the stochastic MPC (sMPC) community [27,28,29] A further method has been proposed by [30] for the case of hard constraints, which constructs a slow non-stationary MDP to promote stability of learning via the implementation of a dynamic penalty method. However, the aforementioned works negate the presence of offline model-process mismatch.…”

A nested online scheduling and nonlinear model predictive control framework for multi‐product continuous systems

“…In chemical process and related industries, the quest for improved economic performance and better operating conditions under an increasingly dynamic market has driven the need to perform integration of multiple decision-making levels. [1][2][3][4][5] It has been broadly recognized that integrating production scheduling and control leads to significant economic benefits as infeasible and suboptimal decisions are diminished. 6,7 A well-known strategy in this area is the inclusion of the process dynamic model into the scheduling framework.…”